Sensitivity and value of information

What drives the answer, and what learning more is worth.

Sensitivity: what drives the answer

A recommendation is only as strong as the estimates behind it. Sensitivity asks which of your estimates, if wrong, would change the answer.

You will meet it in two places:

• inside the report, on each objective (the tornado for that goal), and
• in the Sensitivity analyst tool, which works on value and lets you try what-ifs.

In the report: the tornado for an objective

Open an objective in the results and scroll to Sensitivity analysis.

Each bar is one estimate. The grey side shows the result when the estimate is at its low value (P10), the dark side when it is at its high value (P90), with everything else still uncertain. The black line is the plan’s average and the dashed red line is the target.

Read it like this:

• The longest bars matter most. “Retailer lead conversion” swings revenue growth from about 8% to 16%, far more than “Months to develop the range”.
• A bar that crosses the dashed line means that estimate alone can decide whether you hit the target.
• Put research effort into the estimates at the top. Estimates at the bottom can stay rough.

In the Sensitivity tool

Open Analyst tools, then Sensitivity. It is described as “what moves the answer, what-ifs and stress tests”.

What changes the answer most is the same idea in money. Each bar shows the value of the decision when the estimate is at its low and high value, with the swing written next to it.

Could an estimate change the best option? For each estimate, the table shows which option is best at its low value and at its high value. A highlighted row means the best option flips inside that estimate’s own range. Here, “Months to develop the range” is the one that flips from “Go as planned” to “Go with mitigations”, which is a reason to look harder before committing.

Try a different estimate has a slider for each of the top estimates. Moving one scales that estimate and its range. Nothing is saved until you name and save a scenario, so it is safe to explore. Save a stress test reruns the full simulation with one estimate changed (for example “half as much again”) and saves it next to the base run.

Tool	Question it answers	How to read it
Tornado chart	Which estimates move the result most?	The longest bars matter most
Break-even	How wrong must this estimate be before another option wins?	Far outside your range means the answer is robust to it. Inside your range means that estimate is critical
Stress tests	What if a whole scenario goes against us?	See Stress tests
What-if sliders	What happens if I nudge this?	A quick, unsaved preview. Anything you quote must come from a saved run

Value of information

Before committing, ask whether it would be worth finding out more first. Value of perfect information (EVPI) is how much better, on average, the decision would be if you could learn the true value of every “what we do not know yet” estimate before choosing. It is the most research could ever be worth.

EVPI compared with the cost of finding out	What to do
Very small, or zero	Decide now
Larger than the cost of a quick, cheap check	Consider a targeted study on the top tornado estimates, then re-run
Large, but any study would be slow	Weigh the value of learning against the cost of waiting

Keep in mind:

• It is a ceiling, not a budget. A real study is imperfect, costs money and takes time, so it is worth less than EVPI.
• It counts only what research can reveal. Variation that will happen anyway is excluded.
• Zero is a useful answer. If one option wins in nearly every future, learning more cannot change the decision.

To pick which estimates to research, combine both: research the ones that sit near the top of the tornado and have a break-even inside their range.

The value of a specific study (EVSI) and of learning about a single estimate (EVPPI) are not calculated; only the overall ceiling is.

Next: Revise